Artificial Intelligence is the next big thing, logistics has always been a big thing. So why aren’t they often mentioned together?
Modern manufacturing and logistics systems are supported by increasingly ubiquitous and powerful computing networks. Within these networks, oceans of data are continuously being generated by sensors, machines, systems, smart devices, and people. Together with rising computational capabilities, this Big Data is being analysed faster, more broadly, and more deeply than ever before. These advances have redefined the value of Artificial Intelligence (AI) technologies and opened a new age known as Industry 4.0 or the Smart Factory.
Advanced cognitive computing and deep learning methods have begun to find application in manufacturing systems for automated visual inspections, fault detection, and maintenance. There are active efforts to apply reinforcement learning methods to material handling systems and production scheduling. Industries hoping to convert real-time data into actionable decisions seek opportunities to integrate AI methods with traditional Operational Research approaches, the concepts and technologies of the Internet of Things (IoT), and cyber-physical systems.
The aim of this special issue is to provide insight into the latest advances emerging in the production research community that seek to exploit AI methods in this field. The special issue received 61 submissions from which nine papers were selected. The papers are grouped into three categories: AI methods for manufacturing systems, AI developments specifically in semiconductor manufacturing, and AI in additive manufacturing and maintenance. Within these categories the papers are ordered alphabetically by the last name of the first author.
As part of Industry 4.0, the exploitation of data for intelligent decision making is essential for general manufacturing systems. The paper by Deng et al. considers milling systems and endeavours to predict machining parameters that will provide reliable chatter-free milling. They use a neural network to model the limiting axial cutting depth as a component of the second-order fourth-moment method. The next paper by Fang et al. considers the prediction of the time remaining for jobs to be completed in a job shop. Using big data, they develop a deep learning method for the prediction that is superior to previous regression or network-based predictions in their numerical experiments. The third paper in this category by Rao et al. focuses on a bi-objective welding shop scheduling problem. The problem is modelled as a mixed-integer programme and solved via a non-dominated sorting genetic algorithm with a restarting strategy. Numerical experiments show that the proposed algorithm dominates four other algorithms.
There is much excitement regarding the potential for AI in semiconductor manufacturing. In Chien et al., deep reinforcement learning is exploited to guide the selection of demand forecast models for semiconductor components and modules. An empirical study and a real-world implementation demonstrated the validity of the approach. The identification and classification of defects are essential in semiconductor manufacturing and the paper by Kim et al. studies this important problem. They use a generalised uncertain decision tree model to classify defect patterns on multiple wafer maps based on uncertain features and show that their approach is significantly more efficient than existing methods to analyze real DRAM wafers. The problem of dynamic dispatching for unreliable machines in re-entrant production systems, typically semiconductor manufacturing systems, is considered in the paper by Wu et al. They combine a deep neural network model and Markov decision processes (MDP) to rapidly generate near optimal dynamic control policies for problems that are too large to be only solved by MDP, thus showing the potential of machine learning in controlling unreliable manufacturing systems.
Additive manufacturing and data-driven maintenance are of increasing importance to support the emergence of truly Smart Factories. In the paper by Elhoone et al., a framework for a cyber additive manufacturing system is developed. Three artificial neural network algorithms are proposed and embedded in a two-stage model to support the dynamic allocation of digital designs to different additive manufacturing techniques. The paper by Stanisavljevic et al. provides insights from an experimental study on methods to detect interferences, and thus improve product quality, in real time in additive manufacturing. Their approach is of practical importance and relies on sensor data and methods such as feature selection and machine learning to reach very high detection rates. With an eye toward optimal maintenance decisions in a risk-based environment with fuzzy parameters, the paper by Wang et al. considers belief propagation in constrained fuzzy Bayesian networks. They develop an approach, using a max-min programming model, to address the inference problem. Their framework is validated on a gas compressor maintenance problem.